Sam Altman says OpenAI is testing new features by throwing 'a lot of compute' at them

In a recent announcement, OpenAI's CEO Sam Altman revealed that the company is set to roll out new, compute-intensive features in the coming weeks. Shared via a post on X, Altman indicated that due to the high costs associated with these features, they will initially be accessible only to Pro subscribers, with some new offerings likely incurring additional fees. Altman emphasized that this initiative is an experimental endeavor aimed at pushing AI infrastructure to its limits. He stated, "We also want to learn what's possible when we throw a lot of compute, at today's model costs, at interesting new ideas." This move illustrates OpenAI's ongoing challenge of making advanced AI technology widely available while also managing the substantial expenses tied to computational power. The CEO reassured users that OpenAI's goal is to significantly reduce the cost of AI services over time, promising, "We are confident we will get there over time." Despite the ambitious plans, a request for further comments from OpenAI went unanswered. OpenAI has previously expressed its relentless demand for computing resources. Kevin Weil, the company’s chief product officer, mentioned on a recent episode of the "Moonshot" podcast that any GPUs they acquire are immediately put to use. He remarked, "The more GPUs we get, the more AI we'll all use," drawing a parallel to how increased bandwidth facilitated the growth of video content. In July, Altman announced plans to expand their GPU fleet to over 1 million by year-end, humorously urging his team to work on enhancing that number significantly. For context, Elon Musk's xAI recently unveiled its use of a supercluster named Colossus, which consists of over 200,000 GPUs for training purposes. Other major tech firms, like Meta, are also competing aggressively in this space, with CEO Mark Zuckerberg highlighting their strategy to leverage compute resources as a competitive edge.